Calibration of GEDI footprint aboveground biomass models in Mediterranean forests with NFI plots: A comparison of approaches

Resea ch a icle Calib a ion o GEDI oo p in abo eg ound biomass models in Medi e anean o es s wi h NFI plo s: A compa ison o app oaches Ad i´ an Pascual a,* , Paul B. May b , Aa ´ on C´ a denas-Ma ínez c , Juan Gue a-He n´ andez d , Neha Hunka a , Jamis M. B uening a , Sean P. Healey e , John D. A ms on a , Ralph O. Dubayah a a Depa men o Geog aphical Sciences, Uni e si y o Ma yland, College Pa k, MD, USA b Depa men o Ma hema ics, Sou h Dako a School o Mines and Technology, Rapid Ci y, SD, USA c Depa amen o de Geog a ía Física y An´ alisis Geog ´ a ico Regional, Uni e sidad de Se illa, 41004, Se ille, Spain d Fo es Resea ch Cen e, School o Ag icul u e, Uni e si y o Lisbon, Ins i u o Supe io de Ag onomia (ISA), Tapada da Ajuda, 1349-017, Lisboa, Po ugal e US Fo es Se ice Rocky Moun ain Resea ch S a ion, Ogden, UT, 84401, USA ARTICLE INFO Keywo ds: Spacebo ne lida GEDI calib a ion Biomass modelling In en o y plo s ABSTRACT Obse a ions om he NASA Global Ecosys em Dynamics In es iga ion (GEDI) p o ide global in o ma ion on o es s uc u e and biomass. Foo p in -le el p edic ions o abo eg ound biomass densi y (AGBD) in he GEDI mission a e based on aining da a sou ced om spa sely dis ibu ed ield plo s coinciden wi h ai bo ne lase scanning su eys. Na ional Fo es In en o ies (NFI) a e a ely used o calib a e GEDI oo p in biomass models because hei sampling and posi ional accu acy p e en accu a e coloca ion wi h GEDI o ALS. This omission can limi he ha moniza ion o ju isdic ional biomass es ima es om NFI’s and GEDI; howe e , he e a e me hods a ailable o imp o e he coloca ion o NFI plo s wi h GEDI oo p in s. Focusing on Medi e anean o es s in Spain, we compa ed di e en app oaches o he colloca ion o NFI and GEDI da a: (i) simula ed wa e o ms om ALS; (ii) nea es -neighbo on-o bi GEDI wa e o ms; and (iii) impu ed GEDI wa e o ms impu ed o NFI plo loca ions using a no el geos a is ical me hod. These me hods a e po en ial solu ions o imp o e he local pe - o mance o biomass models and add ess po en ial local sys ema ic de ia ions be ween GEDI and NFI es ima es. We assess he ad an ages and limi a ions o hese me hods o locally calib a e GEDI biomass models and quan i y he impac o geoloca ion e o s in e e ence NFI plo da a. The new biomass models om each me hod we e used o p edic oo p in le el AGBD, which we e hen g idded o a p o ince in he No h-Wes o Spain. I was ound ha he impu a ion app oach is no sensi i e o common e o s in NFI plo geoloca ion, bu i can ou pe o m ALS-based simula ion in some cases, highligh ing he bene i o in o ma ion om mul iple GEDI oo p in s p oxima e o NFI plo s o imp o ing biomass p edic ions. This esea ch p o ides use s wi h bench- ma k o a ailable echniques o locally-calib a e GEDI oo p in biomass models. 1. In oduc ion One o he c i ical aspec s o e es ial ecosys ems moni o ing is he es ima ion o abo e-g ound biomass (AGBD) using h ee-dimensional (3D) e ical ege a ion s uc u e me ics (Bas os e al., 2022; Li e al., 2023). Clima e policies, moni o ing p og ams and en i onmen al man- agemen agencies use biomass s ocks and luxes o in o m on he e ec s o clima e change and human dis u bances (Dubayah e al., 2022a; F iedlings ein e al., 2022; Ma e al., 2023). The Global Ecosys em Dy- namics In es iga ion (GEDI) mission has made a ailable billions o biomass p edic ions ac oss empe a e and opical ecosys ems du ing he las ou yea s (Kellne e al., 2022), d ama ically expanding he abili y o es ima e ca bon s ocks and luxes (Pa e son e al., 2019). The GEDI es ima ion o biomass is globally consis en o e e ence es ima es (A ms on e al., 2023) despi e gaps ac oss biomes and con inen s in he GEDI Fo es S uc u e and Biomass Da abase (FSBD) used o build GEDI biomass models (Duncanson e al., 2022), and conside ing he sub- s an ial he e ogenei y in biomass s ocking (B uening e al., 2023; C ocke e al., 2023). GEDI spa ial sampling densi y is unma ched by any exis ing * Co esponding au ho . E-mail add esses: [email p o ec ed] (A. Pascual), [email p o ec ed] (J. Gue a-He n´ andez), [email p o ec ed] (N. Hunka), [email p o ec ed] (J.M. B uening), [email p o ec ed] (S.P. Healey), [email p o ec ed] (J.D. A ms on), [email p o ec ed] (R.O. Dubayah). Con en s lis s a ailable a ScienceDi ec Jou nal o En i onmen al Managemen jou nal homepage: www.else ie .com/loca e/jen man h ps://doi.o g/10.1016/j.jen man.2025.124313 Recei ed 25 Oc obe 2024; Recei ed in e ised o m 20 Janua y 2025; Accep ed 21 Janua y 2025 Jou nal o En i onmen al Managemen 375 (2025) 124313 A ailable online 31 Janua y 2025 0301-4797/© 2025 The Au ho s. Published by Else ie L d. This is an open access a icle unde he CC BY-NC-ND license ( h p://c ea i ecommons.o g/licenses/by- nc-nd/4.0/ ). spacebo ne senso , bu s ill spa se enough ha he chances o spa ial coloca ion be ween GEDI oo p in s and ield measu emen s is minimal. GEDI biomass es ima es a e no ha monized o he Na ional Fo es In- en o ies (NFI) plo -le el da a ha coun ies use o moni o ing and epo ing. A ha moniza ion be ween NFIs and GEDI da a is impo an o unde s and o es biomass s ocking (Knapp e al., 2020; A aza e al., 2022; Hunka e al., 2023 Pascual e al., 2023) and o ailo global GEDI es ima es o local condi ions, pa ing he way o p o ide solu ions o mission, educ ion and e i ica ion (MRV) p ojec s o e e i o ies lacking ac i i y da a and whe e GEDI is he only sou ce o sus ain p o- jec s and o es managemen h ough ime (Nee , 2021; Ma ino and Bau is a, 2022). Ai bo ne lase scanning (ALS) da a, when a ailable and scheduled combined wi h ield in en o ies, is an incumben me hod o calib a e biomass. Indeed, GEDI biomass models we e buil ollowing he app oach (Duncanson e al., 2022; Kellne e al., 2022), bu he app oach elies on ALS and when a ailable, i a ely ma ches he acquisi ion o biomass da a in mos NFI p og ams. A oiding he ALS dependency can make many housands o in en o y plo s ac ionable o biomass calib a ion. The scale o 12.5-m adius oo p in s is subs an- ially smalle han NFI plo s and his exace ba e dispa i ies be ween GEDI es ima es and NFI biomass alues. Despi e scaling misma ches be ween plo s and oo p in s, se e al s a egies ha e been pu sued o calib a e biomass using on-o bi and simula ed GEDI da a. When ALS is a ailable, use s can ec ea e he GEDI biomass model building using NFI biomass as esponse a iable (Pascual e al., 2023) and a single ALS-simula ed oo p in as p edic o in o ma ion. Howe e , ALS-noncon ingen solu ions a e mo e applicable. To da e, wo ALS-independen me hods ha e been applied o calib a e biomass using GEDI. One is he ma ching o plo s o GEDI da a using assignmen me hods such as nea es neighbo (e.g., Zhang e al., 2022; Bullock e al., 2023). Al e na i ely, he e is a no el p obabilis ic impu a ion app oach o in e GEDI canopy heigh a any gi en loca ion p esen ed in May e al. (2024). These wo me hods ely on he dis ibu ion o mul iple GEDI measu emen s o c ea e a pai be ween he bes ma ch o an NFI plo (neighbo s) and o impu e canopy heigh es ima es using oo p in s a ound NFI loca ions (impu a ion). We explo ed he ad an ages and limi a ions o hese biomass calib a ion al e na i es based on using 25-m GEDI oo p in es ima es o canopy heigh as p edic o s. Ou s udy a ea is in Spain, whe e high-quali y ALS su eys and GEDI es ima es a e empo ally-coinciden o an NFI p og am ha is imp o ing geoloca ion accu acy (Pascual e al., 2021; Pascual and Gue a-He n´ andez, 2023). Geoloca ion e o s in GEDI oo p in da a exis (e.g., Roy e al., 2021; Li e al., 2023; Tang e al., 2023; Holcomb e al., 2024). Howe e less a en ion is paid o ield plo s posi ioning accu acy (Johnson and Ba on, 2004; Mau o e al., 2011). Geoloca ion e o s in NFI su eys can be p oblema ic when in e ence elies on emo ely-sensed me ics (Babcock e al., 2018; Ande sen e al., 2022; Pascual e al., 2020; Sil ei a e al., 2023), and hese ha e no been add essed in he ew s udies ha ha e ied o calib a e biomass using GEDI me ics. Fo ins ance, he s udy om Bullock e al. (2023) pai ed GEDI oo p in s o <300 plo s in Pa aguay ha we e <200 m apa . How a GEDI da a o NFI plo s can ange o ensu e imp o emen s in biomass calib a ions equi es mo e analyses (Zhang e al., 2022). Following wi h Bullock e al. (2023), geoloca ion e o s in he in en o y plo s we e no acknowledged despi e p ac ical unce ain ies ela ed o use o non-su ey g ade posi ioning ins umen s unde dense ee canopies. We know enhancing plo geo- loca ions imp o es he ela ionships be ween lase me ics and a ea-based o es a ibu es (Fassnach e al., 2014; Lu e al., 2016; Pascual e al., 2018; Liu e al., 2023). La ge o se s be ween nominal and eal plo posi ions - and edge e ec s om o es agmen a ion - can educe he a ailabili y o sui able NFI da a and comp omise e o s in biomass calib a ion especially o e highly- agmen ed and di e se o - es s such as in he Medi e anean (Pascual e al., 2020). Hence, using as many sca ce ield plo s as possible in calib a ions is undamen al. The wa e o m impu a ion app oach p esen ed in May e al. (2024) ensu es all NFI plo s a e used o calib a e biomass, bu a he cos o some le el o e o s in he impu ed GEDI canopy heigh me ics o use -speci ied loca ions. The app oach uses Bayesian in e ence, wi h aceable e o alues ha can be subsequen ly inco po a ed in o new biomass calib a ion models. T ade-o s be ween da a cha ac e is ics (simula ed e sus on-o bi da a), scale misma ches be ween NFI plo s and sou ces o GEDI canopy heigh , o he quali y o new es ima es mus be hough ully e alua ed be o e gene alizing an app oach o p oduce locally- ailo ed NFI biomass es ima es om spacebo ne GEDI me ics. Using a p o ince in he No h-Wes o Spain wi h con as ing o es managemen egimes as es ing a ea, we implemen ed he h ee ollowing calib a ion s a egies o calib a e GEDI biomass es ima es using NFI da a. i) Use o simula ed GEDI canopy heigh om ALS da a co- egis e ing he NFI loca ions - ollowing GEDI L4A model building; ii) C ea e pai s o nea es neighbo s be ween plo s and on-o bi GEDI da a, and iii) Implemen he wa e o m impu a ion me hod om May e al. (2024). S anda d and enhanced-geoloca ed posi ions o >1000 plo s we e used o disen angle he pe o mance o hese me hods owa ds equen ly-igno ed bu common geoloca ion e o s p esen in NFI da a. Ou benchma king s udy is a compa ison o al e na i es o calib a e global GEDI biomass es ima es o local condi ions - mo e alike o exis ing ield da a and mo e aligned o o es managemen and ope a- ional moni o ing needs. This a ele an esea ch opic wi h mul iple applica ions such as helping NFI p og ams o ill da a gaps, imp o ing he es ima ion o o es ca bon o be op imize h ough managemen , empowe ing manage s wi h me hods o co ec biases and unce ain ies in GEDI es ima es, bu also owa ds he assimila ion o GEDI da a in o MRV p ojec s ha equi es locally-calib a ed es ima es o mee e i i- ca ion s anda ds. 2. Ma e ial and me hods 2.1. S udy a ea The s udy a ea is Leon p o ince in he No h-Wes o Spain. Loca ed in he Sou he n bo de o he Can ab ic moun ains, Leon is he se en h la ges p o ince in Spain and home o emblema ic o es ese es such as The Picos de Eu opa Na ional Pa k (Fig. 1). The 586,000 ha o o es a ea a e domina ed by ha dwood species (71%) ollowed by coni e ous and mixed o es s (26% and 5, espec i ely). Some o he dominan species a e Que cus py enaica (dominan in 36.2% o he o es a ea), Pinus syl- es is (10.9%) o spa se Que cus ilex (7.3%). The p o ince is o es - s uc u ally ich and species-di e se: he Spanish NFI (SNFI) quan i ies 39 o es ypes, om low-land spa se oak o es s and es o a ion o es s, o dense pine s ands and shade- ole an species such as old-g ow h beech and ches nu o es s in moun ainous a eas. The p o ince is a sui able o disen angle he pe o mance o GEDI biomass es ima es conside ing clima e zones, ansi ion a eas, p esence o moun ain old- g ow h o es s and la ge ex ensions o coni e o es s om pas e o - es a ion e o s. As o oday, only one model in he GEDI L4A p oduc is scoped o all hese o es ypes (Kellne e al., 2022). Recen ALS su - eys and coinciden ield measu emen s o known and enhanced geo- loca ion (Pascual e al., 2021; Pascual and Gue a-He n´ andez, 2023; Hunka e al., 2023) c ea e an op imal scena io o imp o e o es man- agemen in o ma ion h ough biomass calib a ions using GEDI canopy heigh da a and high-quali y NFI da a. 2.2. Na ional o es in en o y da a The SNFI is designed as concen ic plo s anging up o 25-m adius ollowing a 1-km g id o he sys ema ic sampling design A. Pascual e al. Jou nal o En i onmen al Managemen 375 (2025) 124313 2 Fig. 1. The p o ince o Leon in NW Spain is he s udy egion o he esea ch. The NFI sampling g id whe e plo geoloca ion has been imp o ed is show in he map o e he Spanish Fo es Map used o e ie e he main ee species. Scenes o well- ep esen ed o es s a a in he subse o NFI plo s a e shown. Fig. 2. Dis ibu ion o geoloca ion o se co ec ion in SNFI-4 plo s used o he s udy. We showed h ee examples in di e en o es ypes using a lida -based canopy heigh map in he backg ound. A. Pascual e al. Jou nal o En i onmen al Managemen 375 (2025) 124313 3 (´ Al a ez-Gonz´ alez e al., 2014, (Gue a-He n´ andez e al., 2021). Species o ee s uc u al measu emen s a e pa o he ee- and s and-le el a ay o a iables moni o ed documen ed including biomass, which is es ima ed using SNFI species-speci ic allome ies (Pascual and Gue - a-He n´ andez, 2023). We used he la es su ey a ailable o Leon p o ince collec ed be ween Augus –No embe 2019. The e a e 1360 plo s a ailable bu o his esea ch we used 1161 plo s o which upg aded geoloca ion in o ma ion was p o ided by he SNFI head- qua e s. Co ec ing he geoloca ion o se ook an a e age o 2 h 21’ o each plo . This allowed us o ha e a pai o coo dina es, enhanced-geoloca ed and NFI s anda d NFI posi ioning, which is mo e inaccu a e (Pascual e al., 2020). The a e age geoloca ion o se in he NFI samples was 22.7 m and i was below 50 m o 90% o he plo s, some o which exceeded 100 m (Fig. 2). 2.3. Ai bo ne lida o es ima e biomass and simula e GEDI Ai bo ne lida da a is publicly a ailable in Spain hanks o he Land Su ey Ins i u e (Plan Nacional de O hopho og aphy A´ e ea, PNOA). Poin cloud densi y inc eased om 1 o 2 poin s m −2 in ea ly PNOA su eys o ~5 poin s m −2 in ecen p ojec s, imp o ing quali y in 3D s uc u al mapping and he sui abili y o hese su eys o suppo ope a ional o es managemen . The schedules o he SNFI and PNOA ai bo ne lida su eys usually a e no synch onized, bu Leon p o ince is an excep ion. The ALS mapping using he RIEGL LMS-Q1560 senso was conduc ed in 2021 om June 21s o Oc obe 21s in lea -on condi ions. Pe cen ile heigh dis ibu ions o ai bo ne lida echoes we e compu ed using a 2.5-m heigh -b eak o il e ou non- o es ege a ion as shown in he 10-m esolu ion canopy heigh map p oduced (Fig. S1). Me ics om ALS poin clouds co- egis e ing well-geoloca ed NFI plo s we e used o calib a e and p edic biomass. The ALS-based biomass map was used o assess he e ec o new calib a ions compa ed o exis ing on- o bi es ima es. De ails on he a ailable 25-m biomass map can be ound in Pascual and Gue a-He nandez (2023) and in he Supplemen- a y (Fig. S2). The ALS poin cloud da a was used o simula e GEDI canopy heigh me ics a NFI loca ions using he GEDI simula o (Hancock e al., 2019). The ALS da a pass he equi emen s on seasonali y (lea -on) and mini- mum poin cloud densi y (>4 poin s m −2 ) o ensu e quali y in he simula ed GEDI wa e o ms (Hancock e al., 2019). The simula o ope a es by gene a ing wa e o ms o gi en loca ions om disc e e ALS da a and i was used o build GEDI biomass models (Duncanson e al., 2022; Kellne e al., 2022). We simula ed GEDI me ics wice on each plo : o enhanced-geoloca ed loca ions and o unco ec ed posi ions a ailable om p e ious in en o ies. We added wa e o m noise in he GEDI simula ions o mimic on-o bi condi ions. 2.4. Fo es s uc u al me ics and biomass es ima es om GEDI In his sec ion, we desc ibe how we ma ch NFI and GEDI oo p in da a: we used 3 calenda yea s o GEDI da a om 2019 o 2021 on canopy heigh (L2A p oduc , Dubayah e al., 2020) and AGBD es ima es (L4A p oduc , Dubayah e al., 2022b; Kellne e al., 2022). The GEDI da a quali y il e ing was as ollows (mo e de ails in Table S1): wa e- o m ideli y in he selec ed oo p in s exceeded 0.95 and i was g ea e han canopy co e and he algo i hm selec ion in GEDI L2A was op i- mized o each oo p in . Quali y lags in L2A and L4A we e conside ed as well as maximum allowed di e ence o 50 m be ween GEDI ele a ion lowes mode and TanDEM-X e ain ele a ion (K iege e al., 2007) a ailable in he GEDI L2A. Biomass in GEDI is only p edic ed o he lea -on season (L4A quali y lag =1) and his has ele an implica ions o b oadlea o es ypes. A o al o ~236, ~203 and ~239 housand oo p in s passed he il e s (Fig. 3). Fo each pai o NFI loca ions, we sea ched in 25-m in e als o GEDI oo p in s anging 100 m o less om cen e loca ions. Wa e o m acquisi ion ime p o ided in he GEDI L2A p oduc was used o con ol he pai ing o GEDI oo p in s o NFI Fig. 3. Impac o GEDI da a spa si y in he pa ing o NFI plo s o he mos adjacen GEDI oo p in . Da a spa si y is a unc ion on he densi y o GEDI acks, he p opo ion o c oss- acks and he e ec o quali y il e s a educing he numbe o high-quali y GEDI oo p in s along each g ound ack. A. Pascual e al. Jou nal o En i onmen al Managemen 375 (2025) 124313 4 plo s. Unma ched NFI plo s in yea 2019 we e subsequen ly explo ed in 2020 and 2021 un il a ma ch was ound, o o he wise he candida e plo was emo ed om he aining da a. In he 0–25 m dis ance ange be- ween plo s and oo p in s, we compu ed 134 and 148 plo -wa e o m pai s o enhanced and s anda d geoloca ion, espec i ely. Below he 100-m h eshold, we lis ed 538 and 535 da a pai s, which sh inks he aining da a ac ionable o e-calib a ions by ~50% when he op ion is o use nea es -neighbo as in Bullock e al. (2023). The no el app oach on wa e o m impu a ion p esen ed in May e al. (2024), howe e , maximizes all e e ence da a a ailable o GEDI biomass es ima ion. 2.5. Wa e o m impu a ion The wa e o m impu a ion app oach is based on in e pola ion o GEDI spa ially-incomple e obse a ions a ound NFI loca ions and i uses o es masks o accoun o edges in he p ocess (May e al., 2024). The me hod uses a mul i a ia e spa ial model (Taylo -Rod iguez e al., 2019; May e al., 2022) o acknowledge he spa ial co ela ion o GEDI da a be o e gene a ing p obabilis ic p edic ions o GEDI-like canopy heigh me ics a he speci ied NFI geoloca ions. The p edic ed a ibu e o in e es (canopy heigh – GEDI L2A ela i e heigh 98, RH 98 ) is gi en by he expec ed alue o he p edic i e dis ibu ion and he p edic ion unce ain y is gi en by he dispe sion o he dis ibu ion a ound he expec ed alue. The densi y o GEDI oo p in s a ound plo loca ions and he dis ance sea ch o pai RH me ics o plo a ibu es a e impo an ac o s in he impu a ion me hod (May e al., 2024). The impu ed ou - comes a e p edic ed po en ial alues o RH me ics o a gi en NFI plo a he speci ied loca ion and plo size. P edic ion e o s a e inco po a ed in impu ed RH 98 and subsequen ly in o biomass calib a ions ha can be cus omized o ju isdic ions o o es ypes by g ouping NFI da a. We impu ed RH 98 o each pai o NFI geoloca ions accu acies – enhanced and s anda d. 2.6. Local calib a ions o GEDI biomass da a The impac o da a spa si y is mo e e iden when c oss-in e sec ion o GEDI acks is minimal o none-exis ing, leading o la ge da a gaps a ound sampling plo s especially o b oadlea o es ypes a ec ed by biomass quali y il e s (Table S3). By using only lea -on GEDI da a, we na owed he subse o GEDI measu emen s o he in e al be ween Ma ch 16 h o No embe 3 d co esponding o he 2000–2021 median lea -on season pe iod in Leon p o ince. Moun ain o es s in no he n la i udes such as in Picos de Eu opa NP ha e a sho en g owing season: 90 days less app ox. (Capa os-San iago and Rod iguez-Galiano, 2024). Deciduous b oadlea ees (DBT), e e g een b oadlea (EBT), e e g een needlelea (ENT) and deciduous needlea (DNT) a e he ou PFT ha he GEDI mission uses o biomass mapping. The s a a we e g ouped o Eu ope in o one due o he ew GEDI FSBD si es used o build biomass models. One model is used o all biomass p edic ions ac oss Eu opean o es while g assland, sh ubs & woodlands (GSW) ha e hei global model (Table 1). Acquisi ion da e and PFTs a e used o ac ion a lea -on quali y lag ha excludes lea -o condi ions in EBT o DNT o es s and woodlands (71% o Leon p o ince o es a ea is domina ed by b oadlea ed ee species). We used he ollowing h ee sou ces o GEDI canopy heigh as p e- dic o o biomass o wo NFI plo geoloca ion se ies. 1) ALS-simula ed GEDI-alike es ima es o canopy heigh (RH 98 ), 2) wa e o m impu a ion wi h associa ed unce ain ies on impu ed RH 98 alues ollowing he me hod p esen ed in May e al. (2024), 3) pai s o NFI plo s and adjacen GEDI measu emen s o canopy heigh . We used GEDI da a acqui ed du ing he i s h ee yea s o mission (2019/21). These pai s me he ollowing c i e ia: labeled as high quali y acco ding o he GEDI L2A and L4A quali y lag (a combina ion o mul iple wa e o m quali y me ics), less han 100 m om he NFI plo cen e posi ion - hal he sea ch window used in i. e., Bullock e al. (2023) - and undis u bed be ween he ime o NFI measu emen s and he GEDI acquisi ion. Plo s ha es ed o pa ially-ha es ed as in e p e ed om changes in basal a ea and dominan heigh om compa ing consecu i e measu emen s (SNFI-3 e sus SNFI-4) we e emo ed om he aining da a. The associa ed unce ain y (i.e., a iance o he co a ia es) o p e- dic o RH 98 was assumed ze o when using ALS-simula ed (1) and on- o bi GEDI es ima es (3). Fo impu ed RH 98 , we inco po a ed he p e- dic ed a iance in he calib a ion. Fo hese h ee cases, we ained a simple eg ession model (Eq. (1)) be ween squa ed- oo (sq ) ans- o med GEDI RH 98 and e e ence NFI biomass - ollowing GEDI L4A model building app oach he app oach bu using only one p edic o (Table 2; Kellne e al., 2022). The sq -sq ans o ma ion helps induce a linea ela ionship be ween he co a ia e and he esponse.  AGBD √=βo+β1 RH98 +100 √+ ε [Eq. 1] whe e AGBD is he e e ence es ima e in Mg ha −1 o a gi en NFI plo and RH 98 is he GEDI RH 98 impu ed, simula ed o pai ed on-o bi alue o he speci ic plo . We ained a e-calib a ion model o each me hod using SNFI-4 da a o enhanced and o s anda d geoloca ions. Fo es ype-speci ic es ima es o ou egion-dominan ee species well NFI- ep esen ed we e assessed in de ailed. The species included we e Que - cus py enaica (QP, 323 plo s), ac i ely managed Pinus syl es is (PS, 174), Que cus ilex (QI, 68), and Fagus syl a ica ha domina es no he n alley (FG, 87). To assess model pe o mance, we used he absolu e and ela- i e oo mean squa ed e o (RMSE) be ween i ed alues and he NFI biomass da a in he aining se . 2.7. G idded biomass p oduc s The new NFI-calib a ed GEDI es ima es we e applied o on-o bi Table 1 The GEDI L4A biomass models o Eu ope. S a a Model equa ion DBT EU AGBD =0.963 × (−96.531 +7.175 × RH50 +100 √+2.921  RH98 +100 √)2 ENT EU EBT EU DNT EU GSW AGBD =1.118 ×(−124.832 +12.426 × RH98 +100 √)2 No e: EBT, E e g een B oadlea T ees; GSW, G asslands & Sh ubs and Wood- lands; EU, Eu ope: AGBD, abo eg ound biomass densi y (Mg ha −1 ); RH 50 and RH 98 , ela i e heigh 50 and 98, espec i ely, a ailable in he GEDI L2A p oduc . Table 2 Pe o mance o abo eg ound biomass calib a ion models (AGBD, Mg ha −1 ) using on-o bi GEDI, wa e o m impu a ion and simula ion o GEDI me ics using ai bo ne lida . The s anda d de ia ion (SD) o he model e o , he oo mean squa e e o (RMSE) and model bias a e p esen ed o wo classes o geoloca ion accu acies o he NFI plo s used o model building. Geoloca ion accu acy Canopy heigh (RH 98 , m) Model β 0 /β 1 RMSE (Mg ha ¡1 ) RMSE (%) Bias (Mg ha ¡1 ) Enhanced Wa e o m Imp. −101.65/ 1.04 73.05 79.91 −2.46 ALS simula ion −111.60/ 1.13 64.38 70.42 0.00 GEDI L2A −53.26/ 0.58 73.72 82.19 0.00 S anda d Wa e o m Imp. −100.11/ 1.02 73.00 80.18 −2.68 ALS simula ion −15.32/ 0.18 64.46 70.80 0.00 GEDI L2A −45.67/ 0.51 75.47 84.52 0.00 A. Pascual e al. Jou nal o En i onmen al Managemen 375 (2025) 124313 5 oo p in passing quali y il e s o bo h L2 and L4 p oduc s, and collec ed du ing GEDI mission’s i s epoch, om Ap il 2019 o Ma ch 2023. We used hexagonal g ids o summa ize he h ee NFI-calib a ed GEDI biomass dis ibu ions and o c ea e a baseline by g idding on- o bi GEDI da a. Speci ically, we used he H3-scale o 10 (1.5 ha in g id esolu ion, 75.9 m hexagon edge) o size 8 (73.7 ha and 531.3 m edge, espec i ely). Biomass was p edic ed wi hin wo o es ex en masks as o es s ep esen only 37% ou o 15,581 km 2 . We used he s a a used by GEDI mission (PFT) o calib a ion and p edic ion o biomass and class ‘ o es ’ in he 10-m ESA land co e p oduc (Zanaga e al., 2022). We did no conside GEDI oo p in s anging wi hin he GSW class o wi hin land co e classes in he ESA p oduc o he o es a eas. To assess he bene i s o calib a ions owa ds o es managemen planning in he egion we used in e -compa isons o quan ile dis ibu- ions and maps be ween he h ee es ed calib a ion al e na i es and exis ing baselines based on GEDI and ALS-based in e ence. 3. Resul s 3.1. New calib a ions o abo eg ound biomass The abili y o wa e o m impu a ion o de ec changes in he geo- loca ion o NFI plo s had a minimal in luence in he dis ibu ion o canopy heigh es ima es when accoun ing o he co ec ion in he known geoloca ion o se (Fig. 4). Di e ences in RH 98 canopy heigh dis ibu ions we e almos iden ical: as low as 7 cm while o ALS- simula ed GEDI canopy heigh he alue eached 1.40 m o he en i e da ase (1152 plo s). The bias in he p edic ed biomass is highe han o ALS-simula ed me ics and o on-o bi GEDI canopy heigh bu i is a low alue (<5 Mg ha −1 ). Biomass calib a ions using ALS-simula ed GEDI Fig. 4. Dis ibu ion o ela i e squa ed mean e o s (RMSE) and model bias in he es ima ion o abo eg ound biomass densi y (AGBD, Mg ha −1 ) conside ing geoloca ion accu acy in NFI plo loca ions. Compa isons using SNFI-4 da a as e e ence a e p esen ed o on-o bi GEDI, wa e o m impu a ion and simula ion o GEDI me ics using ai bo ne lida (abo e). Geoloca ion o se co ec ion in NFI plo s o e he g adien o canopy heigh da a. The y-axis shows he di e ence be ween ela i e heigh 98 (RH 98 , m) compu ed om lida -simula ed wa e o ms and om he wa e o m impu a ion app oach (below). A. Pascual e al. Jou nal o En i onmen al Managemen 375 (2025) 124313 6 canopy heigh showed he lowes model e o and he mos simila ma ch o he dis ibu ion o e e ence biomass (Table 2). The la ges RMSE was obse ed o he pai ing o GEDI oo p in s o NFI plo s; 10% RMSE compa ed o using ALS-simula ed canopy heigh as model p edic o . 3.2. Fo es ype e ec s and window sea ch h eshold Biomass was modelled o ou ele an o es ypes wi h espec o managemen egimes (>650 plo s, Table 3) and o di e en s ocking pa e ns: mean alues anged om ~30 Mg ha −1 (spa se Que cus oaks) o ~ 200 Mg ha −1 (dense beech s ands in moun ain o es s). E o s in biomass models o spa se oaks (~90% RMSE) ipled he alues o Medi e anean pine o es s ega dless o he calib a ion me hod applied o he geoloca ion accu acy in he aining da a. These esul s o indi- idual o es ypes showed ha simula ed GEDI canopy heigh was no always he mos e icien solu ion in calib a ions. Impu a ion imp o ed he sco es in ~5% om ALS-simula ion o pine o es s (P. syl es is) while he imp o emen eached 15% compa ed o da a pai s be ween adjacen GEDI da a and NFI plo s. Es ima es based on adjacen on-o bi GEDI canopy heigh da a sys ema ically showed he wo s model pe - o mance and highe alues we e obse ed when inc easing he sea ch window o include mo e GEDI oo p in s as candida e o he pai ing o NFI plo s (Fig. S5). In his app oach, he p opo ion o on-o bi GEDI oo p in s passing L4A il e s was lowe o b oadlea ed o es s compa ed o e e g een o es s, which impac s he esul s o impu a ion and he plo - o- oo p in nea es neighbo app oach. Fo ins ance, he RMSE was 30% highe compa ed o impu a ion and ALS-simula ion in Que cus py enaica o es s, he mos dominan o es ype. 3.3. Dis ibu ions o GEDI oo p in and g idded es ima es Calib a ed biomass es ima es emained consis en ega dless o he o es mask used o apply he new models o GEDI on-o bi da a (Fig. S7). Di e ences mos ly occu ed in he 10–40 Mg ha −1 in e al and all dis ibu ions simila ly cap u ed nea ly ze o-biomass condi ions (Fig. S6). The bi-modal shape p edic ions showed lowe maxima o ALS-simula ion compa ed o on-o bi es ima es bo h ope a ional alues (GEDI L4A) and he new calib a ion based on pai ing on-o bi GEDI o NFI samples. Fo wa e o m impu a ion, p edic ions we e simila o ALS- simula ion when using PFTs as p edic ion s a a and o on-o bi da a when using he ESA o es map. The o e all imp o emen om he NFI- based calib a ion was low in he nea es neighbo app oach (Fig. 5). Ou esul s co espond o size 10 o he H3-hexagonal g id, which is ~0.75 he scale o he GEDI L4B p oduc (Fig. S3) and compa able o he g ids we gene a ed. The new calib a ions based on impu a ion and ALS- simula ed canopy heigh es ima es we e mo e simila o biomass es i- ma es om ALS-based in e ence. The e is a esolu ion misma ch be- ween he la e (25 m) and new GEDI biomass g ids, bu i s use o alida ion helps o isualize he co ec ion a ound he 90 h pe cen ile o he dis ibu ion (90–120 Mg ha −1 ). The selec ed masks impac he layou o biomass maps, especially o non- o es and low-biomass condi ions in he cen al sec ions o Leon p o ince (Fig. S9). Di e ences be ween g idded L4A and new calib a- ions we e highe o e moun ainous o es in he No he n ange o he p o ince. The a e age change in mean biomass o he 20,610 hexagons (Fig. S10) showed high mean and median alues o GEDI L4A g idding compa ed o he h ee NFI-calib a ed biomass dis ibu ions i.e., 6.85 and 2.44 Mg ha −1 , espec i ely, o wa e o m impu a ion, o 6.72 and 4.12 Mg ha −1 , espec i ely, when using on-o bi RH 98 as p edic o . Wa e- o m impu a ion showed lowe es ima es (Fig. 6, ed scale) han GEDI L4A p edic ions, while he use o ALS simula ion and on-o bi GEDI p oduced highe alues han he g idded es ima es om GEDI L4A (Fig. 6, blue scale). Opposi e endencies a e especially no o ious in clus e s o e moun ainous o es s and s eep e ain such as in he Leon- side o ‘Picos de Eu opa’ Na ional Pa k whe e biomass es ima es om ALS-based in e ence success ully excluded non- o es pa ches in p e- dic ion s a a. Biomass dis ibu ions in hese a eas looked mo e aligned be ween on-o bi GEDI and impu a ion me hods. The disc epancies o- wa ds he ALS-based simula ion app oach showcased he need o ca e- ully e alua e p edic ions in challenging en i onmen s and also he di e en p ope ies be ween simula ed and on-o bi da a. Biomass ecalib a ions in hese complex en i onmen s wi h espec o opog- aphy and phenology a e especially ele an o he conse a ion and sus ainable managemen o hese old-g ow h high-biodi e si y o es s. 4. Discussion 4.1. Biomass es ima es a plo -le el Ou s udy showcased an in eg a ion o global spacebo ne da a om GEDI in o egional biomass calib a ion o p oduce locally- ailo ed es i- ma es a ju isdic ional le el. Among he es ed me hods, ALS-simula ion p oduced he o e all bes model i ing esul s, bu he app oach is no ope a ional globally as i equi es disc e e lida and assumes ha GEDI wa e o m simula ion e o s a e negligible compa ed o on-o bi GEDI p ope ies. Hence, ALS simula ion should be ega ded as a heo e ical uppe bound o benchma k ope a ional and scalable op ions: wa e o m impu a ion and da a pai s be ween GEDI oo p in s and in en o y plo s based on geoloca ion. Simula ed GEDI canopy heigh om ai bo ne lida yielded he lowes e o when using all NFI plo s a ailable o he 39 o es ypes included in he aining da abase, bu i ing by homogenous g oups o NFI plo s ( o es ypes) esul ed in be e sco es o wa e o m impu a ion on pine s a a – ac i ely managed o imbe p oduc ion - and compa able esul s o o he ha dwood specie. Ou calib a ion me hodologies ha e undamen al p inciples: bo h ALS-simula ion and nea es neighbo app oaches use a single 25-m oo p in wa e o m as p edic o o canopy heigh , while impu a ion uses neighbo ing oo p in s o in e canopy heigh me ics. Wa e o m impu a ion and he NFI- o-GEDI neighbo app oach ely on he dis i- bu ion and densi y o GEDI me ics a ound speci ic a ge s, and his can be mo e in o ma i e o biomass dis ibu ion compa ed o elying on a single well-geoloca ed small-size oo p in o summa ize he s uc u al condi ions o a conside ably la ge plo . The misma ch in scale be ween esponse and explana o y a iables exis : we calib a ed 50-m oo p in biomass es ima es using 25-m Table 3 Pe o mance o abo eg ound biomass calib a ion models (AGBD, Mg ha −1 ) by means o he pe cen ual oo mean squa ed e o (RMSE) using on-o bi GEDI, wa e o m impu a ion and simula ion o GEDI me ics using ai bo ne lida in i e dominan o es ypes. The s anda d de ia ion o he model e o , he oo mean squa e e o (RMSE) and model bias a e p esen ed o wo classes o geoloca ion accu acies o he NFI plo s used o model building. Fo es Type RMSE (%) Mean AGBD (Mg ha ¡1 ) NFI plo s Ra io GEDI sho s/Fo es a ea Impu ed RH 98 Simula ed RH 98 On-o bi RH 98 Enhanced S anda d Enhanced S anda d Q. py enaica 53.63 56.78 67.04 89.22 89.00 64.41 323 1.08 P. syl es is 28.84 33.72 38.17 50.91 51.59 75.32 174 1.22 Q. ilex 89.76 100.16 99.45 107.55 110.32 29.93 68 1.32 F. syl a ica 44.03 39.76 41.65 47.94 46.20 190.1 87 0.67 A. Pascual e al. Jou nal o En i onmen al Managemen 375 (2025) 124313 7 oo p in GEDI da a. Despi e scaling inconsis encies, he calib a ion o biomass by o es ype was e ec i e o lowe es ima ion e o s as shown in p e ious s udies (B uening e al., 2023; C ocke e al., 2023). We ound ha , ega dless o he app oach, biomass calib a ion was a om op imal o low-s a u e spa se o es s (e.g., Que cus ilex), whose ho i- zon al s uc u al condi ions p esen a challenge o he GEDI signal p ocessing especially when canopy co e is low and much o he g ound wi hin he 25-m oo p in is exposed (Dubayah e al., 2020; Do ado-R- oda e al., 2021; Li e al., 2023). Conside ing spa se o es s a e no well sui ed o he a ea-based es ima ion o s ocking (Pascual e al., 2023; Gue a-He n´ andez e al., 2024), impu a ion equaled he pe o mance o ALS-simula ion acco ding he low bias obse ed. The spa se oak o es s we e alua ed a e e e g een and had highe co e age o GEDI da a compa ed o o he b oadlea ed o es s assessed ha we e mo e a ec ed by phenology condi ions and il e ing c i e ia. The phenology lag used in GEDI p oduc s lowe ed he p opo ion o oo p in s ac ionable o biomass calib a ions in b oadlea ed o es s when using on-o bi da a. Resul s o wa e o m impu a ion migh ha e been e en be e i we added mo e obse a ions: da a lagged using canopy heigh (L2A p oduc ) ins ead o he mo e- es ic i e biomass lag (L4A p oduc ). Highe GEDI co e age o e e g een o es s compa ed o b oadlea ed ees es ic ed he ull capaci ies o wa e o m impu a ion as he densi y and he dis ibu ion o GEDI oo p in s play a ole (May e al., 2024). The GEDI mission uses bo h lea -o and lea -on da a o canopy heigh p oduc s bu o consis ency wi h GEDI L4A models, we applied he phenology lag used o GEDI L4A p oduc . This can be e isi ed o i.e., small ju isdic ions whe e da a es ic ion can limi imp o emen s in biomass es ima es, o o speci ic s a a ha manage- men iden i ies as p io i ies. We d opped >50% o on-o bi canopy heigh es ima es o he b oadlea ed o es ypes assessed, bu a mo e local scales in no he n la i udes and highe ele a ions, e.g., Picos de Eu opa NP, he alue is highe as he g owing seasons sho ens compa ed o lowe ele a ions whe e mos o he ac i e o es manage- men (sil icul u e) in he egion akes place, mainly o e pine plan a- ions in ma u e s ages and e ol ed o i egula s and s uc u es. We calib a ed biomass a ju isdic ional le el o each me hod, bu we could ha e s acked p edic ions o indi idual o es ypes p io i ized by managemen needs and economy in he egion (i.e., inco po a ing pine-domina ed a eas o mixed o es s) o by eco egions as shown by Bullock e al. (2023) in Pa aguay o by B uening e al. (2023) in he Uni ed S a es. Indeed, we could ha e agg ega ed o es ypes using ewe classes a highe NFI hie a chies (i.e., genus) o ensu e enough aining da a exis in Leon p o ince i.e., by genus o by age class as in Jha e al. (2024). Howe e , mos o he ep esen ed o es ypes had <30 samples o enhanced geoloca ion, and ha is he eason we p esen biomass map a ju isdic ional-le el o he h ee al e na i es. Fu he esea ch will expand his biomass calib a ion e o na ionwide and o he en i e Ibe ian Peninsula o e he nex yea s making use o incoming ALS acquisi ion in Spain and in Po ugal – he i s coun y-wide su ey in he coun y. 4.2. Biomass p edic ions and g idding Ou oo p in -le el biomass calib a ions we e applied o o es -only a eas using a land co e p oduc ha is accu a e o o es s and ba e- ea h opog aphy, bu less accu a e o sh ublands and g asslands (Zanaga e al., 2022). Mixed condi ions be ween o es s, sh ublands and g asslands o e moun ainous a eas and in he p oximi y o idges can p oduce un ealis ic high es ima es o GEDI canopy heigh (Liu e al., 2021), bu he e a e mo e ac o s o assess he p esence o ou lie s in biomass maps (Kno e al., 2023). To lowe he impac o hose GEDI ou lie s o e sensi i e land co e ypes, he GEDI L4B p oduc assigns ze o mean and unce ain y o hese oo p in s anging o e ba e-ea h e ain using he same p oduc we used (Dubayah e al., 2022a; A m- s on e al., 2023). Manage s in e es ed in u ning ac ionable biomass es ima es and ecalib a ed alues om GEDI should ca e ully e alua e he dis ibu ion o he da a as hese en i onmen s p esen challenging condi ions o canopy heigh mapping (Plane Labs, 2024). A he p edic ion s age, we ound biomass calib a ions based on ALS- simula ion p oduced highe alues, on a e age, compa ed o GEDI L4A g idded es ima es (Fig. S11), which was he opposi e as o impu a ion and on-o bi da a. P ope ies be ween eal and simula ed GEDI da a a e di e en and hese can igge unexpec ed di e ences in he dis ibu ion and alues o biomass es ima es in inal maps as obse ed using quan ile compa isons (Fig. S8): di e ences no cap u ed by he i ed biomass- heigh calib a ions a plo / oo p in le el. The quan ile dis ibu ion o biomass o wa e o m impu a ion and o ALS-simula ion emained simila below he 90 h quan ile. Howe e , a mo e local scales and in p esence o mixed land co e ypes ha lowe he p opo ion o o es s wi hin a gi en ju isdic ion such as wi hin he bounda ies o Picos de Eu opa NP, biomass es ima es can disag ee mo e as he new calib a ion models we e o es -speci ic. Fo ins ance, a subs an ial p opo ion o change es ima es we e abo e 100 Mg ha −1 in Picos de Eu opa NP bu o di e en sign: models om ALS-simula ion abo e-p edic ed while he impu a ion app oach p oduced lowe a e age es ima es compa ed o g idded GEDI L4A baseline. 4.3. Suppo o NFI p og ams and conse a ion inance The spa si y o GEDI acks a ound in en o y plo s a ec s bo h he NFI- o-GEDI pai ing app oach and he wa e o m impu a ion. The la e ensu es all in en o y da a a e used in biomass calib a ions a he cos o Fig. 5. Dis ibu ion quan iles o he h ee biomass calib a ions s a egies compa ed o a 25-m biomass map buil wi h ai bo ne lida (ALS) and NFI da a. A. Pascual e al. Jou nal o En i onmen al Managemen 375 (2025) 124313 8 some p edic ion e o in he canopy heigh p edic o (May e al., 2024). Using nea es neighbo wi h a 100-m maximum sea ch o da a pai ing hea ily educed he aining da a a ailable (>50%). Wa e o m impu- a ion me hod lacks sensi i i y owa ds geoloca ion e o s in he e e - ence plo s and his has mo e p os han cons om he op ics o biomass calib a ions and MRV ac i i ies. Geoloca ion e o s a e conside able in scale and dis ibu ion in many NFI p og ams and en i onmen al moni- o ing ne wo ks (Pascual e al., 2018; Ande sen e al., 2022). The scale o geoloca ion o se s in many o es sampling p og ams is likely la ge han he 23-m o se a e age in Leon p o ince, especially whe e NFI p og ams a e in ea ly s ages and/o whe e he e is limi ed access o high-end ield posi ioning equipmen (Sil ei a e al., 2023). Fig. 6. Change map be ween h ee me hods o gene a e newly-calib a e es ima es o abo eg ound biomass densi y (AGBD, Mg ha −1 ) g idded a ~ 500-m g id compa ed and o sub ac ed om GEDI L4A g idding a he same scale. The p edic ion s a um o apply he new biomass calib a ion equa ions was class ‘Fo es s’ in he ESA V200 landco e p oduc . A. Pascual e al. Jou nal o En i onmen al Managemen 375 (2025) 124313 9